eigen: Return type "annotation" does not reflect actual shape for vectors?

Issue description

Per docs here (as of 2.5.0):
https://pybind11.readthedocs.io/en/stable/advanced/cast/eigen.html#vectors-versus-column-row-matrices

When returning an Eigen vector to numpy, [...] if it is a compile-time vector–that is, the type has either the number of rows or columns set to 1 at compile time–pybind11 converts to a 1D numpy array when returning the value. [...]

Makes total sense. However, looking at the docstrings:
https://github.com/pybind/pybind11/blob/4d9024ec71f30223fc161b7a01fcc486abce800b/tests/test_eigen.py#L609-L623

For m.double_col, the input type makes sense, because that matters to C++ (as mentioned in the docs).

However, the return type is a bit misleading: it states that the output types is numpy.ndarray[numpy.float32[m, 1]].

This is a bit misleading, because the return type's shape is actually (m,).
Additionally, when the return type has

This bit us in https://github.com/RobotLocomotion/drake/issues/13885.
The docs do have a suggested workaround, which is what we'll do, but it'd be nice if the annotation were a bit more truthful.

Reproducible example code

See aforementioned example.

Possible solutions

1. Somehow distinguish the type string based on return-type? (that could be down by "rerouting" the return type somehow?)
2. Discard shape information (at the cost of sacrificing info in the argument - probably a non-starter)
3. As-is
4. Other?

Start with the Eigen vector-versus-column-row-matrix documentation and the referenced section of tests/test_eigen.py around lines 609-623, especially the double_col docstring. Reproduce the mismatch between the return annotation and the actual NumPy shape, then determine which proposed behavior is appropriate and verify the resulting annotation against the example.